LaPIG framework uses Large Language Models to generate paired thermal and visible facial images, addressing the data scarcity problem in facial recognition systems.

• Creates synthetic, high-quality paired datasets without costly data collection
• Combines diffusion models with LLM capabilities for cross-modal image generation
• Enables development of more robust facial recognition systems that work in varying conditions
• Offers significant applications for biometric security and surveillance technologies

This innovation matters for security applications by providing the paired data needed to train systems that can identify individuals in low-light or adverse conditions, enhancing reliability of biometric authentication systems.

LaPIG: Cross-Modal Generation of Paired Thermal and Visible Facial Images